I appreciate that the AI preferentially targets glass cannons, and will minimally attempt to avoid interposing tank units to get to the glass cannon. However, once  intercepted, no attempt will be made by the AI to navigate around the intercepting unit, even if the target is only one square away and can be attacked the same turn. Human opponents will do that, especially to take down troublesome mages and glass cannons. I'd like to recommend a two-part solution to this

1. Implement attacks of opportunity when leaving a threatened square. This will make it harder for human controlled units to reach the juicier targets, and make Krax's fortified squares a little scarier, especially to human players fighting an AI Krax. Since AI will still tend to fight on human players' terms, Krax's fortified squares are already quite effective when used by a human player. This also works well in combination with the movement limit that prevents moving through more than one threatened square. A moderate accuracy or attack power penalty on AoO's will help prevent this interception strategy from being too strong.
2. Give the AI the ability to edge their way around the intercepters towards juicier targets. I realize that implementing AoO's in addition to this further complicates things because the AI has to decide whether it's worth it to risk the extra attacks, but I think it's worth the extra effort: another way human controlled armies take advantage of the AI is to move their units to the back of the map to give archers that extra turn to rain death before moving into melee, and also to wait until the AI army moves within range of a melee charge so the human army can get first strike. The disadvantage of this is there is less space to move, so if the AI can edge around intercepting tank units, they will be right next to the human army's ranged glass cannons.